This is the accepted version of the journal article: Chen, Bingfeng; Zhang, Zhenyan; Zhang, Qi; [et al.]. «Antimicrobial peptides in the global microbiome : biosynthetic genes and resistance determinants». Environmental Science & Technology, Vol. 57, issue 20 (2023), p.7698-7708. DOI 10.1021/acs.est.3c01664 This version is available at https://ddd.uab.cat/record/284315 under the terms of the license 1 Antimicrobial peptides in the global microbiome: Biosynthetic genes and 2 resistance determinants 3 4 Bingfeng Chen1, Zhenyan Zhang1, Qi Zhang1, Nuohan Xu1, Tao Lu1, Tingzhang Wang2, 5 Wenjie Hong2, Zhengwei Fu3, Josep Penuelas4,5, Michael Gillings6, Haifeng Qian1,* 6 7 1. College of Environment, Zhejiang University of Technology, Hangzhou 310032, 8 P. R. of China. 9 2. Key laboratory of microbial technology and bioinformatics of Zhejiang Province, 10 Hangzhou 310012, P. R. of China. 11 3. College of Biotechnology and Bioengineering, Zhejiang University of Technology, 12 Hangzhou 310032, P. R. of China. 13 4. CSIC, Global Ecology Unit CREAF-CSIC-UAB, Bellaterra, Barcelona 08193, 14 Catalonia, Spain. 15 5. CREAF, Campus Universitat Autònoma de Barcelona, Cerdanyola del Vallès, 16 Barcelona 08193, Catalonia, Spain. 17 6. ARC Centre of Excellence in Synthetic Biology, Faculty of Science and 18 Engineering, Macquarie University, NSW 2109, Australia 19 20 21 22 23 *Correspondence to: Haifeng Qian (hfqian@zjut.edu.cn) 24 Abstract: Antimicrobial peptides are a promising new class of antimicrobials that 25 could address the antibiotic resistance crisis, which poses a major threat to human 26 health. These peptides are present in all kingdoms of life, but especially in 27 microorganisms, where they have multiple origins in diverse taxa. To date, there has 28 been no global study on the diversity of antimicrobial peptides, the hosts in which these 29 occur, and the potential for resistance to these agents. Here we investigated the diversity 30 and number of antimicrobial peptides in four main habitats (aquatic, terrestrial, human, 31 and engineered), by analyzing 52,515 metagenome-assembled genomes. The richness 32 of antimicrobial peptides was higher in the human gut microbiome than in other habitats, 33 and most hosts of antimicrobial peptides were habitat-specific. The abundance of genes 34 that confer resistance to antimicrobial peptides varied between habitats and was 35 generally at a low level, except for the built environment and on human skin. The 36 horizontal transfer of potential resistance genes among these habitats was constrained 37 by ecological barriers. We systematically quantified the risk of each resistance 38 determinant to human health and found almost one third pose a risk, especially those 39 that confer resistance to polymyxin B, bacitracin, and defensin. Our results help to 40 identify the biosynthetic potential of antimicrobial peptides in the global microbiome, 41 further identifying peptides where there is a low risk of developing resistance. 42 . 43 Keywords: antibiotic; antimicrobial peptide; resistance; metagenome-assembled 44 genomes; health risk; biosynthesis 45 The prevalence of antibiotic-resistant pathogens exacerbates difficult-to-treat infections 46 in the clinic and poses a serious threat to human health. Infection-related deaths owing 47 to drug-resistant pathogens are projected to account for the highest proportion of deaths 48 worldwide by 2050. Many currently available antibiotics are ineffective due to the 49 development of antibiotic resistance, leading to a crisis in our ability to control bacterial 50 infections (1, 2). Designing and producing novel effective antimicrobial agents is 51 therefore urgently needed. 52 Antimicrobial peptides (AMPs) are attractive antibacterial drug candidates for 53 overcoming the crisis of antibiotic resistance, due to their high diversity and broad 54 activity (3). AMPs are relatively short, structurally diverse small proteins that usually 55 have amphipathic and cationic host-defense properties against pathogen invasion. They 56 are found in diverse taxa, from mammals to soil microorganisms and aquatic 57 environments (4, 5). Thousands of AMPs have been identified from plant, animal, and 58 microbial sources (6), and dozens of AMPs are being evaluated against important 59 pathogens in clinical trials (7–12). AMPs operate by multiple mechanisms, such as 60 membrane disruption, intracellular osmosis, and immune regulation (13). For instance, 61 positively charged AMPs interact with negatively charged substances such as the 62 phospholipids and lipopolysaccharides in bacterial membranes to form lesions and 63 cause cell lysis (5). 64 AMPs are present in all kingdoms of life. Nearly 40% of AMPs in clinical trials, 65 however, are of human origin (3), raising legitimate concerns that the evolution of 66 resistance against AMPs may lead to collateral resistance to endogenous human 67 immunity, which would increase the threat to human health. Therefore, it would be 68 prudent to mine AMPs from other domains of life, and especially from microorganisms, 69 where there is immense phylogenetic and functional diversity. 70 The very ubiquity of AMPs in the environment potentially provides strong 71 selective pressure to drive resistance (5, 14–17), even though the pharmacodynamics 72 and killing mechanisms of AMPs are less conducive to the evolution of resistance than 73 traditional antibiotics (3). However, similarly to antibiotic resistance, antimicrobial 74 peptide resistance genes (AMPRGs) could be acquired by new hosts through horizontal 75 gene transfer (18). This does greatly increase the chance of pathogens potentially 76 acquiring these genes. Thus, before AMPs are used to replace antibiotics in clinical use 77 (19), we should assess the potential threats, screen and develop low threat AMPs to 78 avoid a similar fate to the antibiotic resistance crisis. 79 The identification of genes that encode AMPs has largely been facilitated by 80 studying cultivable microorganisms. However, most microbes are not yet cultivated, 81 and therefore the potential for AMP biosynthesis in most microorganisms remains 82 largely unknown. Further, assigning these genes to their respective hosts remains 83 challenging. In this work, to avoid cultivation bias, we characterized the potential for 84 AMP biosynthesis in microbes at a global scale by analyzing 52,515 metagenome- 85 assembled genomes (MAGs) constructed by Nayfach et al (20) in diverse habitats. By 86 mining these metagenomic data, we discovered widespread AMPs in global 87 microbiomes. Next, we annotated the MAGs with AMPRGs to systematically analyze 88 their distribution patterns and identify their hosts at the global scale. This allowed us to 89 quantify the threat of AMPRGs to human health via three indicators (human 90 accessibility, mobility, and human pathogenicity). These data will guide the 91 development of low health threat AMPs to address the growing antibiotic crisis. 92 93 Results 94 Global distribution patterns of AMPs and their hosts 95 We conducted AMP mining in global microbiome data via the combination of Attention 96 (ATT), long short-term memory (LSTM), and Bidirectional Encoder Representations 97 from Transformers (BERT) models constructing by Ma et al (21). We only keep the 98 results that the accuracy of the three models is greater than 50% suggested by the above 99 reference. Herein, a total of 4,917 AMPs with a length distribution ranging from 21 to 100 50 amino acids in 4,697 non-redundant MAGs were identified to explore the global 101 distribution patterns of AMP biosynthetic potential in microbes (Table S1 and Fig 1b). 102 These MAGs were classified into six habitat categories including air, aquatic, terrestrial, 103 human, engineered, and other hosts (Fig. 1a and Table S1). We used only four main 104 habitats (aquatic, terrestrial, human, and engineered, mainly the built environment) in 105 the subsequent analysis because they covered 90% of samples. 106 In all, 4,364 AMPs were identified from four main habitats based on deep learning 107 predictions. Of four habitats, the human gut had the highest richness of AMPs (Fig. 1c). 108 This conforms to predictions that this habitat has many potential AMP families that are 109 yet to be intensively studied (22). 110 Determining the hosts of AMPs is helpful in mining AMPs with potential 111 applications. Therefore, we identified the AMP hosts by taxonomically annotating each 112 MAG (see “Methods”) and then explored the distribution of these hosts across diverse 113 habitats. A great diversity of hosts carried AMPs in various habitats. A total of 85 phyla 114 were identified across all MAGs, but a more limited number of bacterial phyla were 115 dominant (Fig. 1d). The composition of the hosts of AMPs differed greatly between the 116 four habitats at the phylum level. Firmicutes, common in the human gut microbiota 117 (23), accounted for a large proportion of hosts in human-associated habitats (53.21%), 118 further reflecting the widespread distribution of AMPs in the human gut. Proteobacteria 119 dominated in engineered (27.71%), aquatic (24.80%), and terrestrial (20.97%) habitats. 120 The hosts of AMPs in aquatic and terrestrial environments were more diverse than those 121 in engineered and human habitats, perhaps as a consequence of higher species diversity 122 in these habitats, allowing more diverse sources of AMPs (24). Lachnospiraceae was 123 the main host of AMPs (Fig. 1e), which is abundant in the unperturbed adult microbiota 124 and promotes colonization resistance against intestinal pathogens (25), pointing to this 125 family as a potentially source of AMPs, suggesting that AMPs might participate in 126 pathogen inhibition. The distribution pattern of most hosts of AMPs indicated habitat 127 specificity, with only a few hosts shared between two or more habitats. 128 129 Figure 1. Global distribution patterns of antimicrobial peptides (AMPs) and their 130 hosts. a, Geographical distribution of samples with AMPs, showing richness in various 131 habitats. The size of the dots represents the richness of the AMPs at each sampling site 132 and the colors represents different habitats. The number of samples is in parentheses. 133 Sampling sites without clear geographic coordinates were represented as 0° latitude and 134 0° longitude. b, Length distribution of 4,917 AMPs from the metagenome-assembled 135 genomes (MAGs) in our study. d, Composition of AMP hosts in the four main habitats 136 at the phylum level. e, Number of MAG in each family (most common 20). f, Venn 137 network of shared and specific AMP hosts at the species level in various habitats. 138 Global patterns of genes conferring resistance to AMPs 139 A total of 129 genes predicted to confer resistance to AMPs (AMPRGs) were identified 140 from 4,434 non-redundant MAGs. These predictions were based on the AMP resistance 141 gene database we constructed. This database identified eight different types of 142 resistance phenotypes (Fig. 2a and Table S2; see “Methods”). Of the 129 AMPRGs, 76 143 were predicted to confer resistance to only one class of AMP, while 53 AMPRGs were 144 predicted to confer resistance to multiple AMP classes. 145 The abundance of AMPRGs varied across habitats and was generally low in all 146 sub-habitats, except for the built environment and on human skin (Fig. 2b). The 147 abundance of AMPRGs was highest on human skin, but the number of AMPRGs per 148 metagenome sample in the human digestive system was higher than that on human skin 149 (p < 0.001, one-way ANOVA test; Fig. 2c). In addition, the high abundance of AMPRGs 150 in the built environment (mainly urban subways) (Fig. 2b) could increase the possibility 151 of AMPRG transmission, in a situation similar to the dissemination of ARGs (26). 152 Given that several types of AMPRG might confer resistance to the same AMP or a 153 single AMPRG might confer resistance to multiple AMPs, we summed the relative 154 abundance of AMPRGs in each AMP category for each sample. This revealed an 155 uneven abundance distribution among the various AMPRG types (Fig. 2d). AMPRGs 156 conferring resistance to multiple AMPs were at highest abundance in sub-habitats, 157 except that AMPRGs for colistin and polymyxin B were dominant in both the built 158 environment and on human skin (Fig. 2d). Use of AMPs in these two environments 159 could raise the threat of AMPRGs becoming widespread and potentially cause AMP 160 failure during treatment of infections. For all 129 identified AMPRGs, almost a quarter 161 were specific to one habitat, while one third of AMPRGs were shared among all four 162 main habitats. These latter AMPRGs were mainly predicted to confer resistance to 163 multiple AMPs and polymyxin B (Fig. 2e). 164 Because there was a high abundance and diversity of AMPRGs in habitats 165 dominated by humans, we divided all samples into groups of high and low population 166 density based on the threshold of 58 people per square kilometer (27) to explore the 167 impact of anthropogenic activities on the distribution of AMPRGs. Almost all 168 AMPRGs could be detected in the high population density environments. These were 169 mainly AMPRGs conferring multiple or polymyxin resistance. Only one AMPRG was 170 specific to the low population density environment (Fig. 2f, h). The abundance of 171 AMPRGs in the high-density environments was significantly higher than that in the 172 low-density environments (p < 0.001, two-tailed Welch’s t-test; Fig. 2g). These results 173 suggest that anthropogenic activities drive increases in the abundance and relative 174 diversity of the AMPRGs. 175 176 177 Figure 2. Global distribution patterns of antimicrobial peptide resistance genes 178 (AMPRGs). a, Geographic distribution of samples with AMPRG abundance in various 179 habitats. The size of the dot represents the average abundance of the AMPRGs at each 180 sampling site and the colors represent different habitats. The number of samples is 181 shown in parentheses. Sampling sites without clear geographic coordinates were 182 represented as 0° latitude and 0° longitude. b to d, Abundance (b), number (c), and 183 proportion (d) of AMPRGs grouped by sub-habitats. Abundance was calculated as 184 Reads Per Kilobase per Million mapped reads (RPKM). Different letters indicate group 185 differences between groups determined using a one-way ANOVA test (p < 0.05). e, 186 Upset plot of shared and specific AMPRGs in the four main habitats. The circle 187 represents the classification of AMPRGs shared by the four habitats. f, Venn plot of 188 shared and specific AMPRGs in low- or high-intensity human activities. g, The 189 abundance of AMPRGs in low- or high-intensity human activities. The p value 190 represents the statistical significance (two-tailed Welch’s t-test). h, The classification 191 of shared and specific AMPRGs in low- or high-intensity human activities. 192 193 Distribution of AMPRGs host and mobile genetic elements (MGEs) 194 After taxonomic profiling, 27 bacterial phyla carrying AMPRGs were identified across 195 all MAGs (Table S2). Gammaproteobacteria were the main carrier of AMPRGs, unlike 196 AMPs, in all habitats (Fig. 3a), with Enterobacteriaceae carrying the most AMPRGs, 197 followed by Staphylococcaceae (Fig. 3b). Some AMPRG types tended to be carried by 198 specific bacteria, showing host-specific patterns at the species level. For example, the 199 host of bacitracin resistance genes were almost always Enterobacteriaceae. 200 Enterobacteriaceae also carried other types of AMPRGs, indicating that this taxon was 201 an important reservoir of AMPRGs, much as it is also an important reservoir of 202 antibiotic resistance. Most AMPRG-carrying hosts were unique to particular habitats, 203 especially in engineered and human environments, with a few hosts shared by two or 204 three habitats, showing habitat specificity, further illustrated by the large differences in 205 host composition among the 42 shared AMPRGs in the four habitats (Figs. 3c and S1). 206 Only one host was found in all habitats. Moreover, AMPRG hosts in human habitats 207 were more prone to be shared with the engineered environment than with other habitats. 208 Different habitats contained the same dominant host of AMPRGs, suggesting that 209 AMPRGs confer site-specific advantages on hosts. 210 The analysis of MGEs by the MAG approaches is largely invalid because these 211 MGEs can have different coverage and nucleotide compositions, they are challenging 212 to metagenomically bin correctly (28). Therefore, we collected 47,169 bacterial 213 genomes in NCBI database from aquatic, engineered, human, and terrestrial habitats. 214 Finally, a total of 13,091 bacterial genomes were annotated to the AMPRGs and 6,001 215 of them were identified as pathogens (Table S3). Then, we selected 5 kb upstream and 216 downstream around each AMPRGs to determine the most closely related MGEs most 217 likely to be involved in AMPRGs transfer from a bioinformatic perspective (29). A total 218 of 1439 MGEs were identified in 4,808 bacterial genomes, most of which were 219 transposases (918/1439) (Table S4). The terrestrial and engineered habitats possessed 220 many specific MGEs (Fig. S2). Only Gammaproteobacteria, Alphaproteobacteria, 221 Betaproteobacteria, Firmicutes, Bacteroidota, and Acinobacteriota carried MGEs, 222 with Gammaproteobacteria carrying a large proportion (63.4%) (Fig. S3a). Notably, 223 72.7% of AMPRGs transported between bacteria within the Gammaproteobacteria (Fig. 224 3d). The AMPRG hosts in the various habitats were also notably dominated by 225 Gammaproteobacteria (Fig. 3a). The potentially mobile AMPRGs conferring multiple 226 AMP resistance (40.3%) and polymyxin B (26.4%) resistance were dominant in 227 Gammaproteobacteria (Fig. S3b). Of the 129 evaluated AMPRGs, 75 (58.1%) were 228 identified as potentially mobile AMPRGs in diverse habitats, mainly the engineered 229 and human environments (Table S5). 230 Unique environmental conditions in different ecosystems can affect the 231 distribution of MGEs (30, 31), altering the composition of AMPRG-carrying hosts in 232 diverse habitats. The AMPRG hosts in various habitats in our study, however, were all 233 dominated by Gammaproteobacteria; we assumed that prevalence of all evaluated 234 AMPRGs carried by MGEs in all metagenomic samples was less than 25% due to 235 ecological barriers, even though they are carried by more MGEs (Fig. S4). Moreover, 236 the mobile AMPRGs in bacterial genomes were shared between fewer bacterial species, 237 suggesting fewer transfer events per AMPRG (Fig. 3e). A total of 54.5% (54) of mobile 238 AMPRGs were shared in nonpathogens and pathogens (Fig. S5), suggesting that 239 pathogens may acquire AMP resistance by horizontal gene transfer from nonpathogens. 240 241 242 Figure 3. Hosts and mobility of AMPRGs. a, Composition of AMPRG hosts in four 243 main habitats at the phylum level (and classes in Proteobacteria). b, Number of 244 AMPRGs per MAG in each family (most common 20) plotted against a heatmap for 245 the different families of AMPRGs. c, Venn network of shared and specific hosts 246 containing AMPRGs at the species level in various habitats. d, Network representation 247 of the mobile AMPRGs shared between hosts. Node size and numbers indicate the total 248 number of mobile AMPRGs in each host. Straight and curved lines represent genes that 249 were shared between hosts at different phyla (classes in Proteobacteria) and the same 250 phylum (classes in Proteobacteria), respectively. Line thickness represents the number 251 of mobile AMPRGs shared among the hosts. Only Gammaproteobacteria, 252 Alphaproteobacteria, Betaproteobacteria, Firmicutes, Bacteroidota, and 253 Acinobacteriota carried MGEs. e, Unique mobile AMPRGs were transferred between 254 different numbers of species, only values with at least two transferred AMPRGs were 255 shown. 256 257 Determination of the threat of AMPRGs to human health 258 We quantified the threat of each AMPRG to human health based on its associated 259 pathogenicity (proportion of pathogenic hosts), human accessibility, and mobility 260 (number of MGEs) (see “Methods”), in a similar fashion to the threat evaluation of 261 antibiotic resistance genes (27). Indicators of human pathogenicity and accessibility 262 covered more than 75% of the AMPRGs, with more than half being potentially mobile 263 (Fig. 4a and Table S6). Ultimately, 44.96% AMPRGs posed a significant threat to 264 human health based on our evaluation framework (Fig. 4b). We ranked all the AMPRGs 265 according to their threat and divided the AMPRGs with risks > 0 into four categories 266 (Q1: top 25%, Q2: 25-50%, Q3: 50-75% and Q4: bottom 25%). The multiple AMPs 267 and polymyxin B resistance genes had a wide distribution in the four risk ranks (Fig. 268 4c), and the average risks for Human Cathepsin G and multiple AMPs resistance genes 269 were much higher than others (Fig. 4d). We also calculated the threat for each sample 270 in various habitats, and the results indicated that the samples from engineered 271 environments had the highest threat, followed by human habitats (Fig. 4e). 272 273 Figure 4. Health threat evaluation of AMPRGs to human health. a, Number of 274 AMPRGs with risk index (RI) > 0 in each indicator, including human pathogenicity 275 (HP), human accessibility (HA), and mobility (MO). b, 44.96% of AMPRGs had a risk 276 index RI > 0. The AMPRGs with an RI > 0 were partitioned three categories based on 277 the rank of their RI: (Q1: top 25%, Q2: 25-50%, Q3: 50-75% and Q4: bottom 25%). c, 278 The composition of risk AMPRGs in different AMPRG types. d, The average risk of 279 the AMPRGs in each class. e, RI of each sample in the four major habitats. The 280 engineered environments (mainly built environment) had the highest risk of 281 antimicrobial peptide resistance. The RI of each sample was calculated as the 282 combination of abundance and the RI of the AMPRGs. Data were showed as mean ± 283 standard error of mean (SEM). Different letters indicate differences between groups 284 determined by a one-way ANOVA (p <0.05). 285 286 Discussion 287 The rapid spread of resistance to conventional antibiotics in pathogens has led to one 288 of the most urgent global health concerns. The development of novel antimicrobial 289 agents such as AMPs is the most promising component of the defensive strategy against 290 these pathogens. Soil, human, and marine microbiota contain a large number of bacteria 291 competing for nutrients. This competition could drive the evolution of new AMPs and 292 AMPRGs (32). Any of these microbiotas could be reservoirs of potentially useful AMPs 293 (33). 294 Traditional techniques for culturing microbes, however, limit the exploitation of 295 unculturable microorganisms that represent sources of biodiversity of unexploited 296 AMPs (34). We focused on mining the potential for novel AMPs in microbes. 297 Previously, tools like sequence alignment-based BLAST or hidden Markov models for 298 identifying conserved motifs and domains were typical methods for identifying proteins 299 with similar functions. However, these approaches are difficult to work with in the 300 identification of short peptides like AMPs, especially those that lack significant 301 homology. Many machine learning approaches have been developed recently to address 302 this challenge. Ma et al (21) successfully used three different natural language 303 processing (NLP) models (ATT, LSTM, and BERT) for AMP-mining in human gut 304 microbiomes. Herein, we applied the unified pipeline combined with these models to 305 predict AMPs in the global microbiome, resulting in 4,917 AMPs. Please note all AMPs 306 predicted in this study were potential AMPs, so it is necessary to verify their functions 307 in future experiments. The AMPs were identified in diverse habitats, especially in the 308 human digestive system (mainly the human gut). The diversity of AMPs probably 309 reflects the ability of species to adapt to the unique microbial environment of the niche 310 they occupy, as a single mutation can greatly change the biological activity of each 311 peptide. Complex microbial ecosystems such as the human gut are locations where 312 long-term competition and co-evolution could generate antimicrobial compounds such 313 as AMPs (35). Therefore, the human gut microbiome could serve as a source of AMPs 314 that could be used to address the antibiotic resistance crisis. 315 Some AMPs in the current databases have been experimentally verified, but many 316 of the AMPs in these databases are based on prediction by homology. This is a key 317 knowledge gap, and more information is required to understand how differences in 318 sequence diversity affects biological functions of presumptive AMPs, how these AMPs 319 might then exert their activity against different taxa, and the functional mechanisms by 320 which they operate. Therefore, matching AMPs with their corresponding resistance 321 genes is a dilemma. Further study of the functions of presumptive AMPs is therefore 322 necessary. Classical in vitro assays can evaluate AMPs specific functions (3). However, 323 results in vitro assays can be strongly influenced by assay conditions, such as pH, salt 324 concentrations, nutrients and osmotic stress, and might not accurately reflect the AMPs 325 activity in vivo. Synergistic effects among different AMPs in vivo may also lower the 326 possibility of resistance evolving (36). Some AMPs form pores on the bacterial 327 membrane that then enable other AMPs to enter and attack the cells (37), implying that 328 synergistic and efficacious AMP cocktails from several of these structural classes could 329 be deployed. In summary, the function of AMPs should be tested in vivo to more 330 accurately account for the mechanisms of action of AMPs. 331 An unprecedented view of the global distribution of resistance genes for 332 antimicrobial peptides was generated in this study. A total of 129 AMPRGs were 333 identified in 4434 MAGs. By examining various sub-habitats of each habitat, we 334 demonstrated that AMPRGs were most abundant in the built environment and on 335 human skin. Our results indicate that anthropogenic activities have promoted the spread 336 of AMPRGs. AMPRG abundance, however, was generally low in other sub-habitats. 337 It is still unclear why AMPs are widespread in different habitats, yet resistance 338 genes are at low abundance. The pharmacology of AMPs might contribute to a lower 339 likelihood of developing resistance (38). Unlike antibiotics, most AMPs function by 340 modifying bacterial cell membranes (39), and can kill bacteria within minutes (40). 341 This could prevent bacterial growth and accumulation of potential resistance mutations, 342 especially if the action is all or nothing. Such a situation might mean that there is only 343 a very narrow window of sub-inhibitory concentrations where genetic variants can be 344 selected. Additionally, to resist antimicrobial peptide attack, microorganisms might 345 have to alter the lipid composition and/or organization of their cell membranes, which 346 could be very costly. AMP resistance is often generated via nonspecific mechanisms, 347 and so fewer mutational pathways are available by which resistance to AMPs may 348 evolve (41). 349 We also identified the hosts of AMPRGs in diverse habitats. 350 Gammaproteobacteria were the dominant carrier of AMPRGs in all habitats, and the 351 distribution pattern of hosts was habitat-specific at the species level. 352 Gammaproteobacteria also carried the most mobile genetic elements associated with 353 AMPRGs. We reasoned that the lateral transfer of AMPRGs among microorganisms 354 could be constrained by taxonomic barriers. 355 AMPRGs act primarily by modifying cell membranes. Since membrane formation 356 is the result of networks of interacting proteins that are highly interconnected, it might 357 be difficult for lateral gene transfer to modify these complex cellular subsystems (42, 358 43). However, even so, we do not conclude that AMPs in the clinic or for clinical use 359 are resistance free. For example, once the colistin resistance gene became plasmid 360 borne, it has spread rapidly around the globe (44). Consequently, clinical use of AMPs 361 could accelerate resistance dissemination by selecting for complex mobile genetic 362 elements, as has been seen in the antibiotic crisis. 363 We should learn from the current antibiotic crisis, and prioritize the risks posed by 364 known AMPRGs before widespread adoption of AMPs as therapeutic agents. In doing 365 so, we might avoid the premature appearance of resistance to this promising class of 366 antibacterial agents. With this in mind, we quantified the threat of each of the AMPRGs 367 we detected, based on their human accessibility, mobility and pathogenicity associated 368 with their hosts. Near half of the 129 AMPRGs we detected were determined to pose 369 some risk of compromising future AMP therapies. Most of these genes conferred 370 resistance to multiple AMPs and polymyxin B. For these AMPRGs, we could apply 371 AMP cocktails to minimize the effect of resistance selection or adjuvants to maximize 372 the cost of developing AMP resistance (18). 373 Our study provides a view of the AMP biosynthetic potential in the global 374 microbiome, which will be conducive to mining the resources of AMPs and minimizing 375 the risk of the development of resistance to endogenous host defenses. Apparently low 376 levels of AMP resistance and the potential for phylogenetic barriers to horizontal 377 transfer for AMPRGs do not mean that risks to human health will not occur. However, 378 by learning the lessons of antibiotic resistance, we might be able to prevent the 379 appearance of resistance to AMPs in the future. 380 381 Methods 382 Data sets 383 Nayfach et al (20) constructed a database of 52,515 MAGs recovered from 384 environmentally diverse metagenomes. We downloaded these MAGs with permission 385 from https://genome.jgi.doe.gov/GEMs. The database also provided information for the 386 taxonomies and habitats of MAGs. 387 We constructed an AMP resistance genes database, including the genes with 388 experimentally confirmed influence on AMP susceptibility, collected from the 389 references provided by Kintses et al (45), and the genes associated with AMP resistance 390 in the Comprehensive Antibiotic Resistance Database (CARD) (46). The remaining 391 genes in CARD not associated with AMP resistance were used as a comparable data 392 set. 393 We collected 47,169 bacterial genomes in NCBI database from aquatic (12,037), 394 engineered (11,204), human-associated (12,007), and terrestrial (11,921) habitat to 395 evaluate the mobility and human pathogenicity of AMPRGs more accurately. 396 397 Prediction of AMPs in MAGs 398 Ma et al (21) used AMP sequences from four AMP databases, APD (6), ADAM(47), 399 CAMP(48), and LAMP(49), that contained many AMPs from different species as input 400 data and combined several deep-learning-based natural language processing models, 401 whose performance they optimized to design a pipeline for AMP identification. The 402 combination of ATT, LSTM, and BERT models showed the highest precision. 403 Therefore, we used the combination of the above three models to identify AMPs in 404 52,515 MAGs. 405 406 Annotation and calculation of the abundance of AMPRGs in the MAGs 407 The AMPRGs were annotated with the AMP resistance genes we constructed with strict 408 parameters (e-value ≤10−10, ≥70% identity at the protein level, and 60% query coverage) 409 using Diamond (v2.0.11.149) (50). Reads in each MAG were mapped to AMPRGs 410 using BWA (v0.7.13) (51) and the unmapped reads were removed using Samtools 411 (v1.3.1) (52). The abundance of AMPRGs in each MAG was calculated as Reads Per 412 Kilobase per Million mapped reads based on the number of mapped reads and length 413 of genes. To identify the hosts of AMPRGs, we taxonomically annotated the MAGs 414 containing AMPRGs using the methods described by Nayfach et al (20). Pathogenic 415 hosts were identified using the A-to-Z database (53). We manually reclassified 416 AMPRGs based on the AMPs to which they confer resistance, including polymyxin B, 417 colistin, class IIa bacteriocins, bacitracin, cathelicidins, defensin and nisin. AMPRGs 418 providing resistance to more than one AMP class were classified as multiple resistance 419 determinants. 420 421 Annotation of the MGEs in the MAGs 422 We extracted 5000 bp (29) upstream and downstream of the AMPRGs identified in the 423 bacterial genomes to annotate the MGEs. We used BLASTN (e-value ≤10−10, identity 424 80%, query coverage 80%) to annotate inserted sequences (ISs) with the ISfinder 425 database (54). We annotated transposases and integrases with the NCBI reference 426 sequences that were clustered using CD-HIT (v4.7) (55), with a threshold ≥90%, an e- 427 value ≤10−10, and minimum amino acid identity of 60% over 60% query coverage (56) 428 using Diamond (v2.0.11.149). 429 430 Acquisition of population data 431 We collected the population density of each sampling site from the SEDAC Population 432 Estimation Service (56) and classified all samples into high (>58 people/km2) and low 433 (≤58 people/km2) population densities based on the global average density (27). We 434 only considered the samples for which clear coordinate information was available. 435 436 Determination of AMPRG threat to human health 437 We quantified the threat of AMPRGs to human health based on three indicators, as 438 calculated in our previous study (27): human accessibility, mobility, and human 439 pathogenicity. The accessibility of AMPRGs to humans was the ability to transfer from 440 the environment to humans, calculated as the average abundance and prevalence of the 441 AMPRGs in habitats associated with humans. AMPRG mobility was considered the 442 ability to transfer between hosts by horizontal transfer, calculated as the sum of the 443 number of MGEs (ISs, transposases, and integrases) detected in each bacterial genome. 444 The pathogenicity of an AMPRG was determined using the A-to-Z database (53) and 445 defined as the proportion of pathogenic hosts containing it. Finally, we ranked all the 446 AMPRGs by their threat to human health and divided the AMPRGs with risks >0 into 447 four categories (Q1, Q2, Q3 and Q4). 448 449 Statistical analysis and visualization 450 The data are presented as means ± standard errors of the means. A one-way ANOVA 451 was performed in IBM SPSS Statistics v25.0.0 (IBM Corp., Armonk, NY, USA). Values 452 were considered to differ significantly at p < 0.05. Two-tailed Welch’s t-tests were 453 performed in Excel Analysis Tools (Microsoft Corporation, Redmond, USA). Maps of 454 the geographic distribution of the samples were constructed using the ggmap and 455 ggplot2 packages in R (v4.2.1). Venn diagrams, upset plots, and heatmaps were 456 generated using the OmicStudio tools at https://www.omicstudio.cn. Venn network 457 plots of shared and specific hosts containing AMPs and AMPRGs in diverse habitats 458 were visualized using Cytoscape (v3.9.0). Other plots were visualized using GraphPad 459 Prism (v9.1.1). 460 461 Data availability 462 All metagenome-assembled genomes used in this study are available in IMG/M portal. 463 The bacterial genomes are available in NCBI database. 464 465 Code availability 466 All the scripts and codes for gene annotation, abundance calculation, and visualization 467 used in this study are available online at https://github.com/bingfeng-micro. 468 469 References 470 1. C. Lepore, L. Silver, U. Theuretzbacher, J. Thomas, D. Visi, The small-molecule 471 antibiotics pipeline: 2014-2018. Nat. Rev. Drug Discov. 18, 739–740 (2019). 472 2. J. O’Neill, Tackling drug-resistant infections globally: final report and 473 recommendations (2016). 474 3. B. P. Lazzaro, M. Zasloff, J. Rolff, Antimicrobial peptides: Application informed 475 by evolution. Science. 368, eaau5480 (2020). 476 4. M. Zasloff, Antimicrobial peptides of multicellular organisms. Nature. 415, 389– 477 395 (2002). 478 5. D. I. Andersson, D. Hughes, J. Z. Kubicek-Sutherland, Mechanisms and 479 consequences of bacterial resistance to antimicrobial peptides. Drug Resist. Updat. 480 26, 43–57 (2016). 481 6. G. Wang, X. Li, Z. Wang, APD3: the antimicrobial peptide database as a tool for 482 research and education. Nucleic Acids Res. 44, D1087–D1093 (2016). 483 7. H. B. Koo, J. Seo, Antimicrobial peptides under clinical investigation. Pept. Sci. 484 111, e24122 (2019). 485 8. X. Xu, R. Lai, The chemistry and biological activities of peptides from amphibian 486 skin secretions. Chem. Rev. 115, 1760–1846 (2015). 487 9. M. Simmaco, G. Mignogna, D. Barra, Antimicrobial peptides from amphibian 488 skin: What do they tell us? Pept. Sci. 47, 435–450 (1998). 489 10. J. M. Conlon, The contribution of skin antimicrobial peptides to the system of 490 innate immunity in anurans. Cell Tissue Res. 343, 201–212 (2011). 491 11. P. Nicolas, D. Vanhoye, M. Amiche, Molecular strategies in biological evolution 492 of antimicrobial peptides. Antimicrob. Pept. II. 24, 1669–1680 (2003). 493 12. J. M. Conlon, M. Mechkarska, Host-defense peptides with therapeutic potential 494 from skin secretions of frogs from the family pipidae. Pharmaceuticals. 7, 58–77 495 (2014). 496 13. A. A. Bahar, D. Ren, Antimicrobial Peptides. Pharmaceuticals. 6, 1543–1575 497 (2013). 498 14. J. S. Gunn, S. S. Ryan, J. C. Van Velkinburgh, R. K. Ernst, S. I. Miller, Genetic 499 and functional analysis of a PmrA-PmrB-regulated locus necessary for 500 lipopolysaccharide modification, antimicrobial peptide resistance, and oral 501 virulence of Salmonella entericaSerovar Typhimurium. Infect. Immun. 68, 6139– 502 6146 (2000). 503 15. S. Stumpe, R. Schmid, D. L. Stephens, G. Georgiou, E. P. Bakker, Identification 504 of OmpT as the protease that hydrolyzes the antimicrobial peptide protamine 505 before it enters growing cells of Escherichia coli. J. Bacteriol. 180, 4002–4006 506 (1998). 507 16. V. Nizet, Antimicrobial peptide resistance mechanisms of human bacterial 508 pathogens. Curr. Issues Mol. Biol. 8, 11–26 (2006). 509 17. J. Z. Kubicek-Sutherland, H. Lofton, M. Vestergaard, K. Hjort, H. Ingmer, D. I. 510 Andersson, Antimicrobial peptide exposure selects for Staphylococcus aureus 511 resistance to human defence peptides. J. Antimicrob. Chemother. 72, 115–127 512 (2017). 513 18. P. K. Jangir, L. Ogunlana, R. C. MacLean, Evolutionary constraints on the 514 acquisition of antimicrobial peptide resistance in bacterial pathogens. Trends 515 Microbiol. 29, 1058–1061 (2021). 516 19. C. D. Fjell, J. A. Hiss, R. E. W. Hancock, G. Schneider, Designing antimicrobial 517 peptides: form follows function. Nat. Rev. Drug Discov. 11, 37–51 (2012). 518 20. S. Nayfach, S. Roux, R. Seshadri, D. Udwary, N. Varghese, F. Schulz, D. Wu, D. 519 Paez-Espino, I.-M. Chen, M. Huntemann, K. Palaniappan, J. Ladau, S. Mukherjee, 520 T. B. K. Reddy, T. Nielsen, E. Kirton, J. P. Faria, J. N. Edirisinghe, C. S. Henry, 521 S. P. Jungbluth, D. Chivian, P. Dehal, E. M. Wood-Charlson, A. P. Arkin, S. G. 522 Tringe, A. Visel, H. Abreu, S. G. Acinas, E. Allen, M. A. Allen, L. V. Alteio, G. 523 Andersen, A. M. Anesio, G. Attwood, V. Avila-Magaña, Y. Badis, J. Bailey, B. 524 Baker, P. Baldrian, H. A. Barton, D. A. C. Beck, E. D. Becraft, H. R. Beller, J. M. 525 Beman, R. Bernier-Latmani, T. D. Berry, A. Bertagnolli, S. Bertilsson, J. M. 526 Bhatnagar, J. T. Bird, J. L. Blanchard, S. E. Blumer-Schuette, B. Bohannan, M. A. 527 Borton, A. Brady, S. H. Brawley, J. Brodie, S. Brown, J. R. Brum, A. Brune, D. 528 A. Bryant, A. Buchan, D. H. Buckley, J. Buongiorno, H. Cadillo-Quiroz, S. M. 529 Caffrey, A. N. Campbell, B. Campbell, S. Carr, J. Carroll, S. C. Cary, A. M. Cates, 530 R. A. Cattolico, R. Cavicchioli, L. Chistoserdova, M. L. Coleman, P. Constant, J. 531 M. Conway, W. P. Mac Cormack, S. Crowe, B. Crump, C. Currie, R. Daly, K. M. 532 DeAngelis, V. Denef, S. E. Denman, A. Desta, H. Dionisi, J. Dodsworth, N. 533 Dombrowski, T. Donohue, M. Dopson, T. Driscoll, P. Dunfield, C. L. Dupont, K. 534 A. Dynarski, V. Edgcomb, E. A. Edwards, M. S. Elshahed, I. Figueroa, B. Flood, 535 N. Fortney, C. S. Fortunato, C. Francis, C. M. M. Gachon, S. L. Garcia, M. C. 536 Gazitua, T. Gentry, L. Gerwick, J. Gharechahi, P. Girguis, J. Gladden, M. 537 Gradoville, S. E. Grasby, K. Gravuer, C. L. Grettenberger, R. J. Gruninger, J. Guo, 538 M. Y. Habteselassie, S. J. Hallam, R. Hatzenpichler, B. Hausmann, T. C. Hazen, 539 B. Hedlund, C. Henny, L. Herfort, M. Hernandez, O. S. Hershey, M. Hess, E. B. 540 Hollister, L. A. Hug, D. Hunt, J. Jansson, J. Jarett, V. V. Kadnikov, C. Kelly, R. 541 Kelly, W. Kelly, C. A. Kerfeld, J. Kimbrel, J. L. Klassen, K. T. Konstantinidis, L. 542 L. Lee, W.-J. Li, A. J. Loder, A. Loy, M. Lozada, B. MacGregor, C. Magnabosco, 543 A. Maria da Silva, R. M. McKay, K. McMahon, C. S. McSweeney, M. Medina, L. 544 Meredith, J. Mizzi, T. Mock, L. Momper, M. A. Moran, C. Morgan-Lang, D. 545 Moser, G. Muyzer, D. Myrold, M. Nash, C. L. Nesbø, A. P. Neumann, R. B. 546 Neumann, D. Noguera, T. Northen, J. Norton, B. Nowinski, K. Nüsslein, M. A. 547 O’Malley, R. S. Oliveira, V. Maia de Oliveira, T. Onstott, J. Osvatic, Y. Ouyang, 548 M. Pachiadaki, J. Parnell, L. P. Partida-Martinez, K. G. Peay, D. Pelletier, X. Peng, 549 M. Pester, J. Pett-Ridge, S. Peura, P. Pjevac, A. M. Plominsky, A. Poehlein, P. B. 550 Pope, N. Ravin, M. C. Redmond, R. Reiss, V. Rich, C. Rinke, J. L. M. Rodrigues, 551 W. Rodriguez-Reillo, K. Rossmassler, J. Sackett, G. H. Salekdeh, S. Saleska, M. 552 Scarborough, D. Schachtman, C. W. Schadt, M. Schrenk, A. Sczyrba, A. Sengupta, 553 J. C. Setubal, A. Shade, C. Sharp, D. H. Sherman, O. V. Shubenkova, I. N. Sierra- 554 Garcia, R. Simister, H. Simon, S. Sjöling, J. Slonczewski, R. S. Correa de Souza, 555 J. R. Spear, J. C. Stegen, R. Stepanauskas, F. Stewart, G. Suen, M. Sullivan, D. 556 Sumner, B. K. Swan, W. Swingley, J. Tarn, G. T. Taylor, H. Teeling, M. Tekere, 557 A. Teske, T. Thomas, C. Thrash, J. Tiedje, C. S. Ting, B. Tully, G. Tyson, O. 558 Ulloa, D. L. Valentine, M. W. Van Goethem, J. VanderGheynst, T. J. Verbeke, J. 559 Vollmers, A. Vuillemin, N. B. Waldo, D. A. Walsh, B. C. Weimer, T. Whitman, 560 P. van der Wielen, M. Wilkins, T. J. Williams, B. Woodcroft, J. Woolet, K. 561 Wrighton, J. Ye, E. B. Young, N. H. Youssef, F. B. Yu, T. I. Zemskaya, R. Ziels, 562 T. Woyke, N. J. Mouncey, N. N. Ivanova, N. C. Kyrpides, E. A. Eloe-Fadrosh, A 563 genomic catalog of Earth’s microbiomes. Nat. Biotechnol. 39, 499–509 (2021). 564 21. Y. Ma, Z. Guo, B. Xia, Y. Zhang, X. Liu, Y. Yu, N. Tang, X. Tong, M. Wang, X. 565 Ye, J. Feng, Y. Chen, J. Wang, Identification of antimicrobial peptides from the 566 human gut microbiome using deep learning. Nat. Biotechnol. 40, 921–931 (2022). 567 22. H. Sberro, B. J. Fremin, S. Zlitni, F. Edfors, N. Greenfield, M. P. Snyder, G. A. 568 Pavlopoulos, N. C. Kyrpides, A. S. Bhatt, Large-scale analyses of human 569 microbiomes reveal thousands of small, novel genes. Cell. 178, 1245-1259.e14 570 (2019). 571 23. M. Arumugam, J. Raes, E. Pelletier, D. Le Paslier, T. Yamada, D. R. Mende, G. 572 R. Fernandes, J. Tap, T. Bruls, J.-M. Batto, M. Bertalan, N. Borruel, F. Casellas, 573 L. Fernandez, L. Gautier, T. Hansen, M. Hattori, T. Hayashi, M. Kleerebezem, K. 574 Kurokawa, M. Leclerc, F. Levenez, C. Manichanh, H. B. Nielsen, T. Nielsen, N. 575 Pons, J. Poulain, J. Qin, T. Sicheritz-Ponten, S. Tims, D. Torrents, E. Ugarte, E. 576 G. Zoetendal, J. Wang, F. Guarner, O. Pedersen, W. M. de Vos, S. Brunak, J. Doré, 577 M. Antolín, F. Artiguenave, H. M. Blottiere, M. Almeida, C. Brechot, C. Cara, C. 578 Chervaux, A. Cultrone, C. Delorme, G. Denariaz, R. Dervyn, K. U. Foerstner, C. 579 Friss, M. van de Guchte, E. Guedon, F. Haimet, W. Huber, J. van Hylckama-Vlieg, 580 A. Jamet, C. Juste, G. Kaci, J. Knol, K. Kristiansen, O. Lakhdari, S. Layec, K. Le 581 Roux, E. Maguin, A. Mérieux, R. Melo Minardi, C. M’rini, J. Muller, R. Oozeer, 582 J. Parkhill, P. Renault, M. Rescigno, N. Sanchez, S. Sunagawa, A. Torrejon, K. 583 Turner, G. Vandemeulebrouck, E. Varela, Y. Winogradsky, G. Zeller, J. 584 Weissenbach, S. D. Ehrlich, P. Bork, MetaHIT Consortium (additional members), 585 Enterotypes of the human gut microbiome. Nature. 473, 174–180 (2011). 586 24. L. Flandroy, T. Poutahidis, G. Berg, G. Clarke, M.-C. Dao, E. Decaestecker, E. 587 Furman, T. Haahtela, S. Massart, H. Plovier, Y. Sanz, G. Rook, The impact of 588 human activities and lifestyles on the interlinked microbiota and health of humans 589 and of ecosystems. Sci. Total Environ. 627, 1018–1038 (2018). 590 25. M. T. Sorbara, E. R. Littmann, E. Fontana, T. U. Moody, C. E. Kohout, M. 591 Gjonbalaj, V. Eaton, R. Seok, I. M. Leiner, E. G. Pamer, Functional and Genomic 592 Variation between Human-Derived Isolates of Lachnospiraceae Reveals Inter- and 593 Intra-Species Diversity. Cell Host Microbe. 28, 134-146.e4 (2020). 594 26. D. Danko, D. Bezdan, E. E. Afshin, S. Ahsanuddin, C. Bhattacharya, D. J. Butler, 595 K. R. Chng, D. Donnellan, J. Hecht, K. Jackson, K. Kuchin, M. Karasikov, A. 596 Lyons, L. Mak, D. Meleshko, H. Mustafa, B. Mutai, R. Y. Neches, A. Ng, O. 597 Nikolayeva, T. Nikolayeva, E. Png, K. A. Ryon, J. L. Sanchez, H. Shaaban, M. A. 598 Sierra, D. Thomas, B. Young, O. O. Abudayyeh, J. Alicea, M. Bhattacharyya, R. 599 Blekhman, E. Castro-Nallar, A. M. Cañas, A. D. Chatziefthimiou, R. W. Crawford, 600 F. De Filippis, Y. Deng, C. Desnues, E. Dias-Neto, M. Dybwad, E. Elhaik, D. 601 Ercolini, A. Frolova, D. Gankin, J. S. Gootenberg, A. B. Graf, D. C. Green, I. 602 Hajirasouliha, J. J. A. Hastings, M. Hernandez, G. Iraola, S. Jang, A. Kahles, F. J. 603 Kelly, K. Knights, N. C. Kyrpides, P. P. Łabaj, P. K. H. Lee, M. H. Y. Leung, P. 604 O. Ljungdahl, G. Mason-Buck, K. McGrath, C. Meydan, E. F. Mongodin, M. O. 605 Moraes, N. Nagarajan, M. Nieto-Caballero, H. Noushmehr, M. Oliveira, S. 606 Ossowski, O. O. Osuolale, O. Özcan, D. Paez-Espino, N. Rascovan, H. Richard, 607 G. Rätsch, L. M. Schriml, T. Semmler, O. U. Sezerman, L. Shi, T. Shi, R. Siam, 608 L. H. Song, H. Suzuki, D. S. Court, S. W. Tighe, X. Tong, K. I. Udekwu, J. A. 609 Ugalde, B. Valentine, D. I. Vassilev, E. M. Vayndorf, T. P. Velavan, J. Wu, M. M. 610 Zambrano, J. Zhu, S. Zhu, C. E. Mason, N. Abdullah, M. Abraao, A. Adel, M. 611 Afaq, F. S. Al-Quaddoomi, I. Alam, G. E. Albuquerque, A. Alexiev, K. Ali, L. E. 612 Alvarado-Arnez, S. Aly, J. Amachee, M. G. Amorim, M. Ampadu, M. A.-F. 613 Amran, N. An, W. Andrew, H. Andrianjakarivony, M. Angelov, V. Antelo, C. 614 Aquino, Á. Aranguren, L. F. Araujo, H. F. Vasquez Arevalo, J. Arevalo, C. Arnan, 615 L. E. Alvarado Arnez, F. Arredondo, M. Arthur, F. Asenjo, T. S. Aung, J. Auvinet, 616 N. Aventin, S. Ayaz, S. Baburyan, A.-M. Bakere, K. Bakhl, T. F. Bartelli, E. 617 Batdelger, F. Baudon, K. Becher, C. Bello, M. Benchouaia, H. Benisty, A.-S. 618 Benoiston, J. Benson, D. Benítez, J. Bernardes, D. Bertrand, S. Beurmann, T. 619 Bitard-Feildel, L. Bittner, C. Black, G. Blanc, B. Blyther, T. Bode, J. Boeri, B. 620 Boldgiv, K. Bolzli, A. Bordigoni, C. Borrelli, S. Bouchard, J.-P. Bouly, A. Boyd, 621 G. P. Branco, A. Breschi, B. Brindefalk, C. Brion, A. Briones, P. Buczansla, C. M. 622 Burke, A. Burrell, A. Butova, I. Buttar, J. Bynoe, S. Bönigk, K. O. Bøifot, H. 623 Caballero, X. W. Cai, D. Calderon, A. Cantillo, M. Carbajo, A. Carbone, A. 624 Cardenas, K. Carrillo, L. Casalot, S. Castro, A. V. Castro, A. Castro, A. V. B. 625 Castro, S. Cawthorne, J. Cedillo, S. Chaker, J. Chalangal, A. Chan, A. I. Chasapi, 626 S. Chatziefthimiou, S. R. Chaudhuri, A. K. Chavan, F. Chavez, G. Chem, X. Chen, 627 M. Chen, J.-W. Chen, A. Chernomoretz, A. Chettouh, D. Cheung, D. Chicas, S. 628 Chiu, H. Choudhry, C. Chrispin, K. Ciaramella, E. Cifuentes, J. Cohen, D. A. Coil, 629 S. Collin, C. Conger, R. Conte, F. Corsi, C. N. Cossio, A. F. Costa, D. Cuebas, B. 630 D’Alessandro, K. E. Dahlhausen, A. E. Darling, P. Das, L. B. Davenport, L. David, 631 N. R. Davidson, G. Dayama, S. Delmas, C. K. Deng, C. Dequeker, A. Desert, M. 632 Devi, F. S. Dezem, C. N. Dias, T. R. Donahoe, S. Dorado, L. Dorsey, V. Dotsenko, 633 S. Du, A. Dutan, N. Eady, J. A. Eisen, M. Elaskandrany, L. Epping, J. P. Escalera- 634 Antezana, C. L. Ettinger, I. Faiz, L. Fan, N. Farhat, E. Faure, F. Fauzi, C. Feigin, 635 S. Felice, L. P. Ferreira, G. Figueroa, A. Fleiss, D. Flores, J. L. Velasco Flores, M. 636 A. S. Fonseca, J. Foox, J. C. Forero, A. Francis, K. French, P. Fresia, J. Friedman, 637 J. J. Fuentes, J. Galipon, M. Garcia, L. Garcia, C. García, A. Geiger, S. M. Gerner, 638 S. L. Ghose, D. P. Giang, M. Giménez, D. Giovannelli, D. Githae, S. Gkotzis, L. 639 Godoy, S. Goldman, G. H. Gonnet, J. Gonzalez, A. Gonzalez, C. Gonzalez- 640 Poblete, A. Gray, T. Gregory, C. Greselle, S. Guasco, J. Guerra, N. Gurianova, W. 641 Haehr, S. Halary, F. Hartkopf, J. J. A. Hastings, A. Hawkins-Zafarnia, N. H. 642 Hazrin-Chong, E. Helfrich, E. Hell, T. Henry, S. Hernandez, P. L. Hernandez, D. 643 Hess-Homeier, L. E. Hittle, N. X. Hoan, A. Holik, C. Homma, I. Hoxie, M. Huber, 644 E. Humphries, S. Hyland, A. Hässig, R. Häusler, N. Hüsser, R. A. Petit, B. 645 Iderzorig, M. Igarashi, S. B. Iqbal, S. Ishikawa, S. Ishizuka, S. Islam, R. Islam, K. 646 Ito, S. Ito, T. Ito, T. Ivankovic, T. Iwashiro, S. Jackson, J. Jacobs, M. James, M. 647 Jaubert, M.-L. Jerier, E. Jiminez, A. Jinfessa, Y. De Jong, H. W. Joo, G. Jospin, T. 648 Kajita, A. S. Ahmad Kassim, N. Kato, A. Kaur, I. Kaur, F. de Souza Gomes Kehdy, 649 V. S. Khadka, S. Khan, M. Khavari, M. Ki, G. Kim, H. J. Kim, S. Kim, R. J. King, 650 K. Knights, G. KoLoMonaco, E. Koag, N. Kobko-Litskevitch, M. Korshevniuk, 651 M. Kozhar, J. Krebs, N. Kubota, A. Kuklin, S. S. Kumar, R. Kwong, L. Kwong, I. 652 Lafontaine, J. Lago, T. Y. Lai, E. Laine, M. Laiola, O. Lakhneko, I. Lamba, G. de 653 Lamotte, R. Lannes, E. De Lazzari, M. Leahy, H. Lee, Y. Lee, L. Lee, V. Lemaire, 654 E. Leong, M. H. Y. Leung, D. Lewandowska, C. Li, W. Liang, M. Lin, P. Lisboa, 655 A. Litskevitch, E. M. Liu, T. Liu, M. A. Livia, Y. H. Lo, S. Losim, M. Loubens, 656 J. Lu, O. Lykhenko, S. Lysakova, S. Mahmoud, S. A. Majid, N. Makogon, D. 657 Maldonado, K. Mallari, T. M. Malta, M. Mamun, D. Manoir, G. Marchandon, N. 658 Marciniak, S. Marinovic, B. Marques, N. Mathews, Y. Matsuzaki, V. Matthys, M. 659 May, E. McComb, A. Meagher, A. Melamed, W. Menary, K. N. Mendez, A. 660 Mendez, I. M. Mendy, I. Meng, A. Menon, M. Menor, R. Meoded, N. Merino, C. 661 Meydan, K. Miah, M. Mignotte, T. Miketic, W. Miranda, A. Mitsios, R. Miura, K. 662 Miyake, M. D. Moccia, N. Mohan, M. Mohsin, K. Moitra, M. Moldes, L. Molina, 663 J. Molinet, O.-E. Molomjamts, E. Moniruzzaman, S. Moon, I. de Oliveira Moraes, 664 M. Moreno, M. S. Mosella, J. W. Moser, C. Mozsary, A. L. Muehlbauer, O. Muner, 665 M. Munia, N. Munim, M. Muscat, T. Mustac, C. Muñoz, F. Nadalin, A. Naeem, 666 D. Nagy-Szakal, M. Nakagawa, A. Narce, M. Nasu, I. G. Navarrete, H. Naveed, 667 B. Nazario, N. R. Nedunuri, T. Neff, A. Nesimi, W. C. Ng, S. Ng, G. Nguyen, E. 668 Ngwa, A. Nicolas, P. Nicolas, A. Nika, H. Noorzi, A. Nosrati, H. Noushmehr, D. 669 N. Nunes, K. O’Brien, N. B. O’Hara, G. Oken, R. A. Olawoyin, J. Q. Oliete, K. 670 Olmeda, T. Oluwadare, I. A. Oluwadare, N. Ordioni, J. Orpilla, J. Orrego, M. 671 Ortega, P. Osma, I. O. Osuolale, O. M. Osuolale, M. Ota, F. Oteri, Y. Oto, R. 672 Ounit, C. A. Ouzounis, S. Pakrashi, R. Paras, C. Pardo-Este, Y.-J. Park, P. 673 Pastuszek, S. Patel, J. Pathmanathan, A. Patrignani, M. Perez, A. Peros, S. Persaud, 674 A. Peters, A. Phillips, L. Pineda, M. P. Pizzi, A. Plaku, A. Plaku, B. Pompa-Hogan, 675 M. G. Portilla, L. Posada, M. Priestman, B. Prithiviraj, S. Priya, P. Pugdeethosal, 676 C. E. Pugh, B. Pulatov, A. Pupiec, K. Pyrshev, T. Qing, S. Rahiel, S. Rahmatulloev, 677 K. Rajendran, A. Ramcharan, A. Ramirez-Rojas, S. Rana, P. Ratnanandan, T. D. 678 Read, H. Rehrauer, R. Richer, A. Rivera, M. Rivera, A. Robertiello, C. Robinson, 679 P. Rodríguez, N. A. Rojas, P. Roldán, A. Rosario, S. Roth, M. Ruiz, S. E. Boja 680 Ruiz, K. Russell, M. Rybak, T. S. Sabedot, M. Sabina, I. Saito, Y. Saito, G. A. 681 Malca Salas, C. Salazar, K. M. San, J. Sanchez, K. Sanchir, R. Sankar, P. T. de 682 Souza Santos, Z. Saravi, K. Sasaki, Y. Sato, M. Sato, S. Sato, R. Sato, K. Sato, N. 683 Sayara, S. Schaaf, O. Schacher, A.-L. M. Schinke, R. Schlapbach, C. Schori, J. R. 684 Schriml, F. Segato, F. Sepulveda, M. S. Serpa, P. F. De Sessions, J. C. Severyn, 685 H. Shaaban, M. Shakil, S. Shalaby, A. Shari, H. Shim, H. Shirahata, Y. Shiwa, R. 686 Siam, O. Da Silva, J. M. Silva, G. Simon, S. K. Singh, K. Sluzek, R. Smith, E. So, 687 N. Andreu Somavilla, Y. Sonohara, N. Rufino de Sousa, C. Souza, J. Sperry, N. 688 Sprinsky, S. G. Stark, A. La Storia, K. Suganuma, H. Suliman, J. Sullivan, A. A. 689 M. Supie, C. Suzuki, S. Takagi, F. Takahara, N. Takahashi, K. Takahashi, T. 690 Takeda, I. K. Takenaka, S. Tanaka, A. Tang, Y. Man Tang, E. Tarcitano, A. 691 Tassinari, M. Taye, A. Terrero, E. Thambiraja, A. Thiébaut, S. Thomas, A. M. 692 Thomas, Y. Togashi, T. Togashi, A. Tomaselli, M. Tomita, I. Tomita, X. Tong, O. 693 Toth, N. C. Toussaint, J. M. Tran, C. Truong, S. I. Tsonev, K. Tsuda, T. Tsurumaki, 694 M. Tuz, Y. Tymoshenko, C. Urgiles, M. Usui, S. Vacant, B. Valentine, L. E. Vann, 695 F. Velter, V. Ventorino, P. Vera-Wolf, R. Vicedomini, M. A. Suarez-Villamil, S. 696 Vincent, R. Vivancos-Koopman, A. Wan, C. Wang, T. Warashina, A. Watanabe, 697 S. Weekes, J. Werner, D. Westfall, L. H. Wieler, M. Williams, S. A. Wolf, B. 698 Wong, Y. L. Wong, T. Wong, R. Wright, T. Wunderlin, R. Yamanaka, J. Yang, 699 H. Yano, G. C. Yeh, O. Yemets, T. Yeskova, S. Yoshikawa, L. Zafar, Y. Zhang, 700 S. Zhang, A. Zhang, Y. Zheng, S. Zubenko, A global metagenomic map of urban 701 microbiomes and antimicrobial resistance. Cell. 184, 3376-3393.e17 (2021). 702 27. Z. Zhang, Q. Zhang, T. Wang, N. Xu, T. Lu, W. Hong, J. Penuelas, M. Gillings, 703 M. Wang, W. Gao, H. Qian, Assessment of global health risk of antibiotic 704 resistance genes. Nat. Commun. 13, 1553 (2022). 705 28. F. Maguire, B. Jia, K. L. Gray, W. Y. V. Lau, R. G. Beiko, F. S. Brinkman, 706 Metagenome-assembled genome binning methods with short reads 707 disproportionately fail for plasmids and genomic islands. Microb. Genomics. 6 708 (2020). 709 29. Y. Che, Y. Yang, X. Xu, K. Břinda, M. F. Polz, W. P. Hanage, T. Zhang, Reply 710 to Partridge et al.: Complementary bioinformatics and experimental approaches to 711 investigate the transfer of AMR genes. Proc. Natl. Acad. Sci. 118, e2108995118 712 (2021). 713 30. L. P. Shaw, K. K. Chau, J. Kavanagh, M. AbuOun, E. Stubberfield, H. S. Gweon, 714 L. Barker, G. Rodger, M. J. Bowes, A. T. M. Hubbard, H. Pickford, J. Swann, D. 715 Gilson, R. P. Smith, S. J. Hoosdally, R. Sebra, H. Brett, T. E. A. Peto, M. J. Bailey, 716 D. W. Crook, D. S. Read, M. F. Anjum, A. S. Walker, N. Stoesser, null null, 717 Niche and local geography shape the pangenome of wastewater- and livestock- 718 associated Enterobacteriaceae. Sci. Adv. 7, eabe3868. 719 31. J. Bethke, A. Davidovich, L. Cheng, A. Lopatkin, W. Song, J. Thaden, V. Fowler, 720 M. Xiao, L. You, Environmental and genetic determinants of plasmid mobility in 721 pathogenic Escherichia coli. Sci. Adv. 6, eaax3173 (2020). 722 32. Arnold Markus F. F., Shabab Mohammed, Penterman Jon, Boehme Kevin L., 723 Griffitts Joel S., Walker Graham C., Genome-wide sensitivity analysis of the 724 microsymbiont Sinorhizobium meliloti to symbiotically important, defensin-like 725 host peptides. mBio. 8, e01060-17 (2017). 726 33. N. J. Tobias, H. Wolff, B. Djahanschiri, F. Grundmann, M. Kronenwerth, Y.-M. 727 Shi, S. Simonyi, P. Grün, D. Shapiro-Ilan, S. J. Pidot, T. P. Stinear, I. Ebersberger, 728 H. B. Bode, Natural product diversity associated with the nematode symbionts 729 Photorhabdus and Xenorhabdus. Nat. Microbiol. 2, 1676–1685 (2017). 730 34. M. Magana, M. Pushpanathan, A. L. Santos, L. Leanse, M. Fernandez, A. 731 Ioannidis, M. A. Giulianotti, Y. Apidianakis, S. Bradfute, A. L. Ferguson, A. 732 Cherkasov, M. N. Seleem, C. Pinilla, C. de la Fuente-Nunez, T. Lazaridis, T. Dai, 733 R. A. Houghten, R. E. W. Hancock, G. P. Tegos, The value of antimicrobial 734 peptides in the age of resistance. Lancet Infect. Dis. 20, e216–e230 (2020). 735 35. B. Vishnepolsky, M. Grigolava, G. Managadze, A. Gabrielian, A. Rosenthal, D. 736 E. Hurt, M. Tartakovsky, M. Pirtskhalava, Comparative analysis of machine 737 learning algorithms on the microbial strain-specific AMP prediction. Brief. 738 Bioinform. 23, bbac233 (2022). 739 36. Yu Guozhi, Baeder Desiree Y., Regoes Roland R., Rolff Jens, Combination effects 740 of antimicrobial peptides. Antimicrob. Agents Chemother. 60, 1717–1724 (2016). 741 37. M. R. Yeaman, N. Y. Yount, Mechanisms of antimicrobial peptide action and 742 resistance. Pharmacol. Rev. 55, 27 (2003). 743 38. G. Yu, D. Y. Baeder, R. R. Regoes, J. Rolff, Predicting drug resistance evolution: 744 insights from antimicrobial peptides and antibiotics. Proc. R. Soc. B Biol. Sci. 285, 745 20172687 (2018). 746 39. A. Rodríguez-Rojas, O. Makarova, J. Rolff, Antimicrobials, stress and 747 mutagenesis. PLoS Pathog. 10, e1004445 (2014). 748 40. G. E. Fantner, R. J. Barbero, D. S. Gray, A. M. Belcher, Kinetics of antimicrobial 749 peptide activity measured on individual bacterial cells using high-speed atomic 750 force microscopy. Nat. Nanotechnol. 5, 280–285 (2010). 751 41. N. Jochumsen, R. L. Marvig, S. Damkiær, R. L. Jensen, W. Paulander, S. Molin, 752 L. Jelsbak, A. Folkesson, The evolution of antimicrobial peptide resistance in 753 Pseudomonas aeruginosa is shaped by strong epistatic interactions. Nat. Commun. 754 7, 13002 (2016). 755 42. R. Jain, M. C. Rivera, J. A. Lake, Horizontal gene transfer among genomes: The 756 complexity hypothesis. Proc. Natl. Acad. Sci. 96, 3801–3806 (1999). 757 43. O. Cohen, U. Gophna, T. Pupko, The complexity hypothesis revisited: 758 connectivity rather than function constitutes a barrier to horizontal gene transfer. 759 Mol. Biol. Evol. 28, 1481–1489 (2011). 760 44. R. Wang, L. van Dorp, L. P. Shaw, P. Bradley, Q. Wang, X. Wang, L. Jin, Q. 761 Zhang, Y. Liu, A. Rieux, T. Dorai-Schneiders, L. A. Weinert, Z. Iqbal, X. Didelot, 762 H. Wang, F. Balloux, The global distribution and spread of the mobilized colistin 763 resistance gene mcr-1. Nat. Commun. 9, 1179 (2018). 764 45. B. Kintses, O. Méhi, E. Ari, M. Számel, Á. Györkei, P. K. Jangir, I. Nagy, F. Pál, 765 G. Fekete, R. Tengölics, Á. Nyerges, I. Likó, A. Bálint, T. Molnár, B. Bálint, B. 766 M. Vásárhelyi, M. Bustamante, B. Papp, C. Pál, Phylogenetic barriers to 767 horizontal transfer of antimicrobial peptide resistance genes in the human gut 768 microbiota. Nat. Microbiol. 4, 447–458 (2019). 769 46. A. G. McArthur, N. Waglechner, F. Nizam, A. Yan, M. A. Azad, A. J. Baylay, K. 770 Bhullar, M. J. Canova, P. G. De, L. Ejim, L. Kalan, A. M. King, K. Koteva, M. 771 Morar, M. R. Mulvey, J. S. O’Brien, A. C. Pawlowski, L. J. V. Piddock, P. 772 Spanogiannopoulos, A. D. Sutherland, I. Tang, P. L. Taylor, M. Thaker, W. Wang, 773 M. Yan, T. Yu, G. D. Wright, The Comprehensive Antibiotic Resistance Database. 774 Antimicrob. Agents Chemother. 57, 3348–3357 (2013). 775 47. D. Veltri, U. Kamath, A. Shehu, Deep learning improves antimicrobial peptide 776 recognition. Bioinformatics. 34, 2740–2747 (2018). 777 48. F. H. Waghu, R. S. Barai, P. Gurung, S. Idicula-Thomas, CAMPR3: a database on 778 sequences, structures and signatures of antimicrobial peptides. Nucleic Acids Res. 779 44, D1094–D1097 (2016). 780 49. X. Zhao, H. Wu, H. Lu, G. Li, Q. Huang, LAMP: a database linking antimicrobial 781 peptides. PloS One. 8, e66557 (2013). 782 50. B. Buchfink, K. Reuter, H.-G. Drost, Sensitive protein alignments at tree-of-life 783 scale using DIAMOND. Nat. Methods. 18, 366–368 (2021). 784 51. H. Li, R. Durbin, Fast and accurate short read alignment with Burrows–Wheeler 785 transform. Bioinformatics. 25, 1754–1760 (2009). 786 52. H. Li, B. Handsaker, A. Wysoker, T. Fennell, J. Ruan, N. Homer, G. Marth, G. 787 Abecasis, R. Durbin, 1000 Genome Project Data Processing Subgroup, The 788 Sequence Alignment/Map format and SAMtools. Bioinformatics. 25, 2078–2079 789 (2009). 790 53. HARTMANN Science Centre, Relevant Pathogens from A–Z https://www.bode- 791 science-center.com/center/relevant-pathogens-from-a-z.html (2021). 792 54. P. Siguier, J. Perochon, L. Lestrade, J. Mahillon, M. Chandler, ISfinder: the 793 reference centre for bacterial insertion sequences. Nucleic Acids Res. 34, D32– 794 D36 (2006). 795 55. L. Fu, B. Niu, Z. Zhu, S. Wu, W. Li, CD-HIT: accelerated for clustering the next- 796 generation sequencing data. Bioinformatics. 28, 3150–3152 (2012). 797 56. Y. Che, Y. Xia, L. Liu, A.-D. Li, Y. Yang, T. Zhang, Mobile antibiotic resistome 798 in wastewater treatment plants revealed by Nanopore metagenomic sequencing. 799 Microbiome. 7, 44 (2019). 800 57. SEDAC, Population Estimation Service https://sedac.ciesin.columbia.edu/ 801 mapping/popest/pes-v3/ (2021). 802 803 Acknowledgements 804 We appreciate Nayfach et al. for their work on the genomic catalog of Earth's 805 microbiomes and its publication as a public resource. This study was financially 806 supported by the National Key Research and Development Program of China 807 (2017YFD0200503), the Key Research and Development Program of Zhejiang 808 (2022C02029 and 2022C02046), and the National Natural Science Foundation of China 809 (21777144, 21777145, and 21976161). JP acknowledge financial support from the 810 Spanish Government grants PID2019-110521GB-I00 and TED2021-132627B-I00, the 811 Fundación Ramón Areces grant CIVP20A6621, and the Catalan Government grant 812 SGR 2017-1005. 813 814 Contributions 815 Bingfeng Chen designed the study with guidance from Haifeng Qian. Bingfeng Chen 816 wrote the first draft of the manuscript and Haifeng Qian, Josep Penuelas, Zhengwei Fu, 817 and Michael Gillings contributed substantially to revision. Tingzhang Wang and Wenjie 818 Hong contributed to the all-functional annotation of metagenome-assemble genomes. 819 Bingfeng Chen and Zhenyan Zhang, and Tao Lu contributed to the collection and 820 disposal of all data information. Bingfeng Chen, Qi Zhang, and Nuohan Xu performed 821 all the metagenomic analysis. Bingfeng Chen and Zhenyan Zhang performed the 822 visualization of all data and the artistic design of all figures. Haifeng Qian, Zhengwei 823 Fu, Tao Lu, and Josep Penuelas acquired funding for this project. 824 825 Ethics declarations 826 Competing Interest 827 The authors declare that they have no competing interests.